Performance Assessment of Genomic Selection on Stripe Rust Across Generations and Environments in Bi-parental Wheat Populations

To improve early-stage selection for stripe rust resistance in wheat, it is essential to use genomic selection models that can effectively capture genetic variation across different generations and environmental conditions. We assessed the performance of GS models for stripe rust resistance in a bi-parental wheat population derived from a cross between a resistant and a susceptible parent, selected to segregate for resistance under field conditions. Using RR-BLUP and Bayesian approaches, we trained models on F₆ lines and validated them on F₇ lines across four environments to evaluate prediction accuracy across generation and environments. Bayesian models showed better prediction accuracy (r = 0.59) than RR-BLUP (r = 0.49) to predict F₇ from F₆. Environment-specific prediction, training in one environment and testing in another—yielded lower accuracy (r = 0.36), underscoring the impact of environmental variability on model performance. Incorporating genotype × environment (G×E) interaction within a multi-environment Bayesian framework moderately improved prediction (r = 0.57) across environment. However, internal cross-validation within the training set produced low predictive accuracy (r = 0.09), suggesting potential overfitting or poor generalizability within the same environment. Our findings demonstrate the prediction potential of Bayesian models for predicting stripe rust resistance across generations and environments in highly structed populations like bi-parental wheat populations. However, to enhance model reliability and ensure broader applicability, further validation through additional testing is necessary. We recommend incorporating replicated trials and expanding the training population with larger and more diverse datasets to better capture environmental variation and improve model robustness.